Breast tumours are known to have high levels of intratumour heterogeneity^1-3. Tumours demonstrating high levels of heterogeneity are of a more ‘aggressive’ phenotype^1-3. Histogram analysis provides a method of characterising lesion heterogeneity via a number of summary statistics (mean, median, standard deviation, skewness, kurtosis, and percentiles)^1-3, yet traditionally, imaging metrics tend to be assessed via mean and median values^1-3.

The purpose of this study is to i) determine if histogram analysis of pre-treatment DCE-MRI parameters are associated with traditional prognostic indicators; ii) to conclude if an assessment of lesion heterogeneity provides superior prognostic information than mean or median metrics; and iii) assess if pre-treatment DCE-MRI vascular kinetics provide independent prognostic information regarding early (≤36months) breast cancer recurrence.

MR imaging was undertaken on a 3.0T scanner (GE Healthcare) prior to neoadjuvant chemotherapy (NAC). In each case a 3D dynamic dataset was acquired utilising VIBRANT with a temporal resolution of ~30secs. Semi-automated ROI’s were generated on each slice that demonstrated malignant tissue throughout the breast to generate a 3D volume of interest (VOI).

For DCE-MRI analysis the signal intensity time course was assessed in a pixel-by-pixel manner across all dynamic phases. Linear interpolation was employed to determine vascular parameters. Histogram analysis of the whole lesion was undertaken to allow an assessment of tumour heterogeneity and resulted in first order statistics (mean, standard deviation, skew, kurtosis, median, 5th, and 95th percentiles) for the following model free empirical parameters: maximum enhancement index, time to maximum, rise time, normalised maximum intensity time ratio, percentage of the maximum enhancement index recorded at 30 seconds (PC₃₀), initial slope, final slope and AUC at 60 seconds (AUC₆₀), see Figure 1.

Clinical records provided the following traditional survival indicators: grade (I and II or III), oestrogen receptor (ER) status (negative or positive), progesterone (PR) status (negative or positive), human epidermal growth factor receptor 2 (HER2) status (negative or positive), molecular subtype (triple negative or all other), T stage (≤T2 or >T2), and N stage (N0 or ≥N1).

To assess significant differences within traditional prognostic indicators non-parametric Mann Whitney U-tests were performed.

For survival analysis patients were categorised as having a critical survival event or censored. Critical events were defined as local tumour recurrence and/or distant metastasis (DFS). Patients without critical events were censored. The DFS time interval was defined as the time from initiation of NAC to critical or censored event. A Cox’s proportional hazards model (CPHM) was used for both univariate and multivariate survival analysis.

To avoid over-parameterisation and increase model generalisation a 2-stage feature selection methodology was utilised. Firstly, only those DCE-MRI parameters that demonstrated a post Bonferroni correction significant (p ≤0.05) Mann Whitney U-test result were entered into the univariate CPHM. Secondly, to avoid over-parameterisation, while allowing a comparison with all traditional prognostic indicators only parameters demonstrating significant univariate results were entered into the multivariate CPHM along with all traditional parameters.

To allow appropriate dichotomisation of DCE-MRI parameters the Youden’s Index⁴ was utilised to highlight a suitable threshold for each DCE-MRI parameter.

208 subjects underwent analysis. Following Bonferroni correction 4 significant differences were revealed based on mean or median metrics whereas 31 significant differences were noted for standard deviation, skew, kurtosis, 5th and 95th percentiles (Table 1). Standard deviation alone represented 16 significant differences (Figure 2).

158 individuals were followed up for 36 months and were entered into the survival analysis. The number of critical and censored events along with median follow-up intervals are presented in Table 2. Significant univariate survival DCE-MRI parameters are presented in Table 3. When considering multivariate DFS analysis the following variables were retained by the CPHM: ER, T-stage and initial slope standard deviation (Table 3).

The low number of significant U-test results for mean and/or median measures suggests that other metrics, particularly standard deviation, might better characterise lesions and provide better opportunities for planning personalised therapy.

Regarding DFS, histogram based metrics revealed that tumours demonstrating greater dispersion of the initial enhancement (PC₃₀, initial slope) and the amount of contrast agent delivered to and retained by the tumour (AUC₆₀) along with higher hot (PC₃₀ 95th percentile) and lower cold (rise time 5th percentile) spot values of the initial enhancement are all traits of shorter DFS intervals. Multivariate survival analysis demonstrated that DCE-MRI histogram parameters provided independent prognostic information supplementing traditional survival parameters.

Pre-treatment DCE-MRI histogram parameters are associated with traditional prognostic indicators, have superior prognostic information than mean and/or median values and provide independent prognostic information regarding early (≤36months) breast cancer recurrence.